TinyTorch/docs/tito/quick-reference.md

⚡ Quick Reference

One-page cheatsheet for experienced developers

Essential Commands

Setup & Verification

# Initial setup
git clone https://github.com/mlsysbook/TinyTorch.git
cd TinyTorch
./setup-environment.sh
source activate.sh

# Verify installation
tito system health

# System information
tito system info

Core Workflow

# 1. Edit module in Jupyter
cd modules/NN_name
jupyter lab NN_name.ipynb

# 2. Export to package
tito module complete N

# 3. Validate (run milestone scripts)
cd milestones/MM_YYYY_name
python script.py

Module Management

# Export specific module
tito module complete 01

# Check module status
tito checkpoint status

# System reset (if needed)
tito system reset

Community & Benchmarking

# Join community (optional)
tito community join

# Run baseline benchmark
tito benchmark baseline

# View profile
tito community profile

Module Dependencies

Foundation (Required for All)

Modules 01-07: Tensor, Activations, Layers, Losses, Autograd, Optimizers, Training

Unlocks:

• Milestone 01: Perceptron (1957)
• Milestone 02: XOR Crisis (1969)
• Milestone 03: MLP (1986)

Architecture (Vision + Language)

Modules 08-13: DataLoader, Convolutions, Tokenization, Embeddings, Attention, Transformers

Unlocks:

• Milestone 04: CNN (1998) - requires M01-09
• Milestone 05: Transformer (2017) - requires M01-13

Optimization (Production)

Modules 14-19: Profiling, Quantization, Compression, Memoization, Acceleration, Benchmarking

Unlocks:

• Milestone 06: Torch Olympics (2018) - requires M01-19

Capstone

Module 20: Torch Olympics Competition

Common Workflows

Starting a New Module

# Navigate to module directory
cd modules/05_autograd

# Open in Jupyter
jupyter lab 05_autograd.ipynb

# Implement required functions
# Run inline tests
# Add docstrings

# Export when complete
cd ../..
tito module complete 05

Debugging Module Errors

# Check system health
tito system health

# View detailed error logs
tito module complete N --verbose

# Reset if corrupted
tito system reset

# Reimport in Python
python
>>> from tinytorch import *
>>> # Test your implementation

Running Milestones

# After completing Foundation (M01-07)
cd milestones/01_1957_perceptron
python 01_rosenblatt_forward.py

# After completing Architecture (M01-09)
cd milestones/04_1998_cnn
python 01_lenet_inference.py
python 02_lenet_training.py

# After completing Optimization (M01-19)
cd milestones/06_2018_mlperf
python benchmark.py

File Structure

TinyTorch/
├── modules/           # Source notebooks (edit these)
│   ├── 01_tensor/
│   │   └── 01_tensor.ipynb
│   └── ...
├── tinytorch/         # Exported package (auto-generated)
│   ├── core/
│   ├── nn/
│   └── ...
├── milestones/        # Validation scripts (run these)
│   ├── 01_1957_perceptron/
│   ├── 02_1969_xor/
│   └── ...
└── tito/              # CLI tool

Import Patterns

# After exporting modules
from tinytorch.core.tensor import Tensor           # M01
from tinytorch.nn.activations import ReLU, Softmax # M02
from tinytorch.nn.layers import Linear             # M03
from tinytorch.nn.losses import CrossEntropyLoss   # M04
from tinytorch.autograd import backward            # M05
from tinytorch.optim import SGD, Adam              # M06
from tinytorch.training import Trainer             # M07

Troubleshooting Quick Fixes

Issue	Quick Fix
ModuleNotFoundError: tinytorch	Run tito module complete N to export
tito: command not found	Run source activate.sh
Import works in notebook, fails in Python	Restart Python kernel after export
Tests pass in notebook, fail in milestone	Check module dependencies (M01-07 required)
OOM errors	Profile memory usage, check for unnecessary copies
NaN losses	Check gradient flow, activation stability

Performance Expectations

Baseline Benchmarks (Your Hardware May Vary)

• M01-07: Perceptron trains in <1 second
• M01-09: CIFAR-10 CNN trains in 10-30 minutes (CPU)
• M01-13: Small transformer inference in 1-5 seconds
• M01-19: Optimized models run 10-40× faster

Memory Usage

• M01: Tensor operations ~100 MB
• M01-09: CNN training ~1-2 GB
• M01-13: Transformer training ~2-4 GB
• M01-19: Optimized models use 50-90% less memory

NBGrader (For Students in Courses)

# If using NBGrader in classroom setting
# Submit your completed notebook
# Do NOT submit the exported package

# Grading components:
# - 70% Auto-graded (tests)
# - 30% Manual (systems thinking questions)

Next Steps

• New to TinyTorch? Start with Getting Started
• Stuck on a module? Check Troubleshooting
• Need detailed docs? See CLI Documentation
• Teaching a course? See For Instructors & TAs

---

💡 Pro Tip: Bookmark this page for quick command reference while building!